Dissemination and Mechanism for the MCR-1 Colistin Resistance

1The polymyxins (polymyxin E (colistin) and polymyxin B) are a family of cationic polypeptide antibiotics with a lipophilic fatty acyl side chain. The initial binding of polymyxins bacterial surface mainly depends on the electrostatic interaction between the positively-charged polymyxin and the negatively-charged phosphate group of lipid A on lipopolysaccharide (LPS) localized on the outer leaflet of the bacterial outer membrane. Following its diffusion from the outer membrane across the periplasm, polymyxin intercalates into the inner membrane and forms pores, which in turn results in bacterial lysis. Although they belong to an old generation of antibiotics, polymyxins represent the last line of defense against lethal infections by gram-negative pathogens with pan-drug resistance. Unfortunately, certain species of the Enterobacteriaceae like K. pneumoniae have been recently showing an appreciable resistance to colistin. Indeed, colistin resistance (i.e., inefficient binding of polymyxins to the lipid A moiety of lipopolysaccharide) is mainly due to the 4’-phosphoethanolamine (PEA) modification of the lipid A on the LPS. This type of chemical modification on the bacterial lipid A can be attributed to either the chromosome-encoded machinery in K. pneumoniae or the plasmid-transferred mobilized colistin resistance (MCR-1) gene in certain species of Enterobacteriaceae like E. coli. For the former, two sets of bacterial two-component systems (pmrAB plus phoPQ) and the regulator mgrB are implicated, in which the lipid A of LPS is chemically modified and thereafter exhibits reduced affinity to polymyxin. The latter represents an unique mechanism for bacterial colistin resistance in that the mcr-1 gene product, annotated as a member of a family of phosphoethanolamine transferases, catalyzes the modification of lipid A moiety on bacterial LPS (Fig 1). To the best of our knowledge, the natural occurrence of the mcr-1 gene has been traced to no less than five species: E. coli, Salmonella enteric, K. pneumonia, Enterobacter aerogenes and E. cloacae (of note, it was also experimentally spread/transmitted from E. coli to Pseudomonas aeruginosa by conjugation.). Also, the range of host reservoirs with potential to carry the mcr-1-harbouring enterobacteria extends from poultry/livestock (chickens, pigs, dogs, and cattle) to humans, and published data from January-April 2016 suggests that the mcr-1 gene has been disseminated into no less than 18 countries. To a certain degree, the global spread of the mcr-1 gene might be related to a food-chain based dissemination pathway, which was shown by Zhu’s group. Thus, they observed the paralleled existence of mcr-1 in meat/food samples and in the healthy human microbiome. Worryingly, the MCR-1 colistin resistance gene was strikingly shown to coexist with other multiple-drug resistance genes (i.e, carbapenem and extended-spectrum β-lactam), highlighting the possibility that micro-organisms with pan-drug resistances are emerging. For instance, a variant of the notorious NDM-1 was detected to coexist with MCR-1 in the Enterobacteriaceae (NDM-5 in K. pneumoniae and NDM-9 in a chicken meat isolate of E. coli). So far, most of the studies in this field focused on epidemiological investigations, which is in part due to the relatively limited availability of the genomic information. Nevertheless, the mechanism for transfer, origin, and biochemical analysis of the diversified plasmid-borne MCR-1 colistin resistance remains poorly understood, and these questions are addressed here, in aiming to close the missing knowledge gap.

2The recent emergence of colistin resistance may be attributed to MCR-1-mediated PEA addition of lipid A moiety at the 4’-phosphate group (Fig 1A), a component of bacterial LPS on the outer layer of outer-membrane for Gram-negative bacteria like E. coli (Fig 1B). Consistent with scenarios seen in Neisseria and E. coli, our MS result suggested that the peak of lipid A with mass of 1797.4 is present in the colistin-susceptible E. coli MG1655 (S1A Fig), and one more peak of PPEA-4’-lipid A with mass of 1919.8 appears upon the arabinose-inducing expression of the MCR-1 enzyme (S1B Fig). It validated that the essence of MCR-1-catalyzed enzymatic reaction is the addition of PPEA (mass: 123 Au) to lipid A (mass: 1797.4–1797.6). Very recently, we successfully isolated mcr-1-harbouring plasmids from the colistin-resistant E. coli strains obtained from the gut microbiota of clinically diarrheal patients. Here we subjected the mcr-1-positive plasmids to genome sequencing by next generation desktop MiSeq sequencer (Illumina). A pool of 350-bp paired-end reads we produced, were assembled with GS De Novo Assembler into two long contigs. We then checked the assembly of our plasmids (pE15004 and pE15017) through integrating raw data, PCR assays, and Sanger sequencing. The plasmid pE15004 was assembled correctly, while a ~2.2 kb fragment was missing in the original assembly of pE15017. Consequently, complete genomes of these two clinical mcr-1-positive plasmids (pE15004, in Fig 2A and pE15017 in Figs 2B and 3A–3D) were acquired.

3The mcr-1-harbouring plasmid pE15004 was 33.309 kb in length with a GC content of 41.8%. This plasmid contained 51 predicted ORFs, among which 11 were associated with the formation of type IV pilus (Fig 2A). The backbone of pE15004 was closely-related to the pir-type E. coli plasmids pSAM7 from cattle in the United Kingdom and pJIE143 from human in Australia (S2 Fig), both of which are narrow-host-range IncX4-type plasmids. Further comparative analysis indicated that plasmid pE15004 was nearly identical to IncX4 mcr-1-harboring E. coli plasmids pICBEC72Hmcr (Accession no.: CP015977) isolated from Brazil and pESTMCR (Accession no.: KU743383) from Estonia, K. pneumoniae plasmids pMCR1_Incx4 (Accession no.: KU761327) from China (Fig 3E) and pMCR1.2-IT (Accession no.: KX236309) identified in Italy (Table 1). Another mcr-1-bearing plasmid pE15017 (65.375 kb) we sequenced contained 91 ORFs (Fig 2B), sharing nearly all its sequences with the first-identified mcr-1-harbouring IncI2-type plasmid pHNSHP45 (Fig 3 and S2 Fig). In comparison with the prototype mcr-1-positive plasmid pHNSHP45, the upstream insertion sequence ISApl1 flanked mcr-1 was consistently missing in both pE15004 and pE15017(Fig 3E) as well as in other recently-reported mcr-1-containing plasmids, like pKH457-3-BE.

4Besides the mcr-1 gene (Fig 3A and 3C), our PCR assays combined with Sanger sequencing determined that pE15017 carries an extended-spectrum β lactamase (ESBL) gene bla CTX-M-55 (Fig 3A and 3B). Thus pE15017 represents an ESBL and MCR-1-coproducing plasmid (Fig 3A, 3D and 3E). Previously, the co-occurrence of ESBL and MCR-1 had been found on the IncHI2-type plasmids from E. coli and Salmonella enterica. Similar to the other mcr-1-carrying plasmid, pA31-12 (Fig 3D), pE15017 might be an additive example of IncI2-type plasmid with above two antibiotic resistance determinants. Interestingly, four base pair (AACA, 1612–1615) is consistently missed in the ISEcp1-bla CTX-M-55-orf477 operon (3090 bp) in both pE15017 and pA31-12, in relative to the counterpart (3094 bp) in pHN122-1(Fig 3D). In fact, the ISEcp1-bla CTX-M-55 transposition unit flanked by DR was also recently observed in the Salmonella plasmid pSCS23 (KU934209) as shown in Fig 3.

5Taken together, these results indicate that plasmid pE15004 is an additive member of the mcr-1-carrying plasmids, while pE15017 together with other recently identified plasmids such as pmcr1_IncI2 (Accession no.: KU761326) and pBA76-MCR-1(Accession no.: KX013540), are recent variants of plasmid pHNSHP45 (Figs 2, 3E and S2 Fig). Furthermore, the mcr-1 gene has been found carried by plasmids belonging to at least 5 different incompatibility groups (Table 1), verifying a trend of diversification. Though it is currently only found in Enterobacteriaceae, its dissemination to broad host range plasmids and subsequent spread to a broad bacterial host range cannot be excluded. Moreover, comparative analysis showed that nearly identical mcr-1-containing plasmids were discovered in different countries, suggesting that, besides the possibility that the dissemination of the mcr-1 gene was captured independently from a common ancestor, the direct spread of bacteria harboring the same plasmid is not impossible.

6Although Liu and coworkers determined that the expression of plasmid-borne mcr-1 confers colistin resistance to certain species of the Enterobacteriaceae family, functional details of the MCR-1 protein are poorly understood. Here, we attempt to address this issue. Philius Transmembrane Prediction Server (http://www.yeastrc.org/philius/pages/philius/runPhilius.jsp) suggested that the MCR-1 protein is an integral membrane protein with five trans-membrane regions (S3A Fig). Similar to the LptA (EptA) of Neisseria, the multiple sequence alignments indicated that the MCR-1 protein also belongs to a family of phosphoethanolamine lipid A (PEA) transferases with putative conserved sites (E246, T285, H395, D465 and H466, in S3B Fig) required for its catalytic activity, i.e., the addition of PEA to lipid A from phosphatidylethanolamine (Fig 1). Because the fact that the nascent LPS in cytoplasm is flipped by the ABC transporter MsbA into periplasm and the covalent modification of the lipid A component on LPS occurs in periplasm, it is speculated that the trans-membrane regions ensures the correct anchoring of the MCR-1 enzyme to the periplasmic face of the cytoplasmic membrane attached to the catalytic domain of PEA transferase. While, experimental evidence for this hypothesis is lacking, we aimed to address them using the integrative approaches ranging from protein biochemistry, bioinformatics and structural biology to bacterial genetics.

7We over-expressed the membrane protein MCR-1 and purified it to homogeneity (S4A Fig) and confirmed this by Western blot using an anti-6x-His primary antibody (S4B Fig). MS-based identification further confirmed the identity of the recombinant MCR-1 trans-membrane protein (S4C Fig). To further gain structural insights into the biochemical mechanism of MCR-1, we applied both protein engineering and structure-guided mutagenesis. In particular, the arabinose-inducible expression system pBAD24/MG1655 was also utilized to probe the above concerns in vivo. Given the fact that i) the chemical modification phosphate group of lipid A at 1 or 4’-position impair its binding to polymyxins, ii) the newly-synthesized LPS is translocated by the MsbA lipid flippase into periplasm from cytoplasm; iii) bacterial periplasm is the only place where the moiety of lipid A on LPS is covalently modified with either 4-amino-arabinose or phosphoethanolamine, it is prerequisite that the enzyme modifier including MCR-1 should be localized in bacterial periplasm. Thus, we are extremely interested in determining physiological role of the trans-membrane region in MCR-1 function.

8Firstly, MG1655 with/without the empty vector pBAD24 (the negative control) fails to grow on the LBA plates with above 2 mg/L of colistin, whereas the positive control, MG1655 with the arabinose-induced expression of mcr-1, can grow well on the solid media with 16–32 mg/L of colistin (Table 2). In contrast, pBAD24-facilitated expression of Neisseria lptA conferred the colistin-susceptible MG1655 strain with an ability to grow on the LBA plates with 8–16 mg/L colistin (Table 2). Though the Neisseria lptA encodes the colistin resistance at an appreciable level, the amplitude of drug resistance is less than that of the MCR-1 (Table 2). Subsequently, we engineered a deletion mutant of the mcr-1 gene (Δtm) whose protein product lacks the N-terminal trans-membrane region to further evaluate its role in vivo. Similar to the scenario with the negative control, we found that the E. coli strain with the araninose-induced expression of the mcr-1 mutant (Δtm) cannot grow on the LBA plates with over 2 mg/L of colistin (Table 2), validating the importance of the transmembrane region in the MCR-1-mediated colistin resistance. Thus it may be concluded that the catalytic activity for PEA transferase depends on its location of MCR-1 on bacterial inner-membrane.

9Using the Neisseria Lipo-oligosaccharide Phosphoethanolamine Transferase A (LptA) as structural template (PDB: 4KAV), structural modeling by Swiss-Model program allow us to visualize the architecture of PEA transferase domain of the membrane-bound MCR-1 enzyme (Fig 4A). Extensive analyses of structural docking together with sequence alignments allow us to hypothesize that the following five residues (E246, T285, H395, D465, and H466) are critical for the substrate binding of MCR-1, and in turn determines the MCR-1-encoded colistin resistance (Fig 4B). Driven by this speculation, we used site-directed PCR mutagenesis to create the following point mutations (E246A, T285A, H395A, D465A, and H466A). In contrast to the positive control carrying the wild-type mcr-1 gene (Fig 4C), none of the MG1655 strains expressing the mcr-1 point mutants were observed to grow significantly on the condition of above 2.0 mg/L of colistin, which is almost identical to that of negative control (Fig 4C and Table 2). This represents in vivo evidence that the five residues are essential for the function of MCR-1.

10A BLASTp search for MCR1 and Neisseria gonorrhoeae LptA (EptA) returned a large set of divergently related sequences (annotated as PE transferases, Sulfatases or membrane proteins. Detailed comparisons of alignment methods applied to divergently related sequences have produced low-accuracy results with sequence identities below 30%. We have thus limited our search scope to 30%. To determine a phylogenetic profile of MCR-1, Multiple sequence alignment of the dataset was performed by MUSCLE (http://www.ebi.ac.uk/Tools/msa/muscle) using default parameters and the quality of the alignment was evaluated using Guidance (http://guidance.tau.ac.il). Consequently, we retrieved 32 candidate proteins that returned hits with >30% identity. Maximum Likelihood (ML) phylogenic trees were reconstructed by using the LG amino acid substitution model with gamma distribution and invariant sites selected using MEGA version 6. To ensure appreciable reliability, the results we obtained were validated by 1000 bootstrap replicates.

11Intriguingly, the reconstruction of a maximum phylogeny tree using 32 unique proteins selected from the BLASTp search allowed us to clearly observe two distinctive clades: one containing a family of PEA transferases including MCR-1 and Neisseria LptA (Fig 5) and the other containing an array of putative sulfatases (Fig 5). Also, the members of the PEA transferase family are divided into two apparent subclades, one of which features MCR-1, and the other one comprising Neisseria LptA (Fig 5). The chromosomally-encoded LptA from Neisseria species is most-closely clustered with putative integral membrane proteins found in other pathogenic Ɣ-proteobacteria, whereas the plasmid-borne MCR-1 on the other hand is very close to PEA transferases from the colistin-producing bacteria, esp. the Paenibacillus species (Fig 5). Despite the fact that MCR-1 and LptA share very low sequence identity to each other and fall into two separate subclades within the tree (Fig 5), they still remain functionally-equivalent (Table 2).

12The phylogenetic tree here indicates a divergent evolutionary pattern between the LptA/MCR-1 integral membrane proteins and other putative sulfatases. A domain analysis of MCR-1 revealed distinct trans-membrane helices followed by a sulfatase domain. Sulfatases that catalyze the hydrolysis of a sulfate group are present in all three domains of life and constitute a heterogenic group of enzymes. Due to similarity in size between a sulfate and a phosphate group, one can easily imagine why PEA transferases share core catalytic features with sulfatases. In fact, closely-related sulfotransferases from Mycobacterium transfer a sulfate group into the glycolpeptidolipids (GPL), the equivalent of the LPS in gram-negative bacteria. Due to lack of sufficient sequence data and experimental validation, it is hard to trace the ancestry of MCR-1 to its chromosomal origins. Given the fact that removal of the trans-membrane region from the MCR-1 protein damages its function of MCR-1-mediated colistin resistance (Table 2), one can speculate that acquisition of a trans-membrane domain could have easily enabled these PEA transferases to correctly localize in the inner membrane and to eventually target a variety of substrates with different implications ranging from cationic antimicrobial peptide resistance in the case of lipid A modification, to changes in motility when FlgG, a flagella rod protein, is modified. Given the fact that phylogenetic tree places the MCR-1 protein very close to the PEA transferases from the Paenibacillus family, the known producers of polymyxins, but in a different sub-clade than Neisseria LptA (Fig 5), it raises the possibility that 1) the cousins of Paenibacillus might be highly relevant to its origin of the MCR-1; 2) a potentially parallel evolutionary path is implicated for the two genes (mcr-1 and lptA) under similar environmental selection pressures, e.g., the massive use of colistin as a veterinary medicine.

13The data we present represents a first comprehensive glimpse of mechanisms for diversified plasmid transfer, evolutionary origin, and catalytic reaction of the MCR-1-mediated colistin resistance. The discovery of new mcr-1-harbouring plasmids (pE15004 and pE15017) adds new knowledge into the newly-emerging field of MCR-1 and colistin resistance, furthering our understanding of the diversity in the dissemination of the mcr-1 gene. Unlike the paradigm mcr-1-positive plasmid pHNSHP45 that is isolated from a swine E. coli in southern China, the two plasmids we reported are extracted from clinical E. coli isolates of diarrhea patients. Further functional definition of plasmid genomes delineated that 1) the plasmid pE15004 is a IncX4 plasmid of around 33kb long, differing from the IncI2-type plasmid pHNSHP45 of about 64kb in length (Figs 2, 3 and S2 Fig); 2) the other plasmid pE15017 (~65kb) seemed to be a recent variant of pHNSHP45 (~64kb) in that an insertion sequence ISApl1 of around 1 kb in front of mcr-1 gene of pHNSHP45 is absent in pE15017 (Fig 3E) and an ESBL encoding gene was captured (Fig 3A, 3B and 3D). In agreement with proposal by Petrillo and coauthors, the deletion of this ISApl1insertion sequence might be the relic of the mcr-1 dissemination. It was reported that the mcr-1 colistin resistance gene is present in a multidrug-resistant plasmid or coexists with other resistance genes like extended spectrum β-lactamase, and even the notorious NDM-1 and its variants (NDM-5 and NDM-9). We also found that the pE15017 is an ESBL and MCR-1 coproducing plasmid similar to pA31-12, a recently-isolated plasmid from China. These facts imply that multi-drug, even pan-drug, resistant bacteria with colistin resistance will eventually evolve, a fact that deserves close attention. By contrast, the pE15017 plasmid carries both ESBL and MCR-1. Consistent with our recent observations with swine lung microbiota, the results highlight the differences amongst the mcr-1-carrying plasmid reservoirs in human gut microbiota.

14We experimentally validated that the expression of Neisseria LptA augments colistin resistance of E. coli (Table 2), despite being weaker than MCR-1 (Table 2), suggesting the possibility of various catalytic aspects and differing evolutionary paths for the two genes (lptA and mcr-1). To address this concern, we conducted phylogenetic analyses and found they are placed into two neighboring sub-clades of the PEA transferase family (Fig 5), giving a strong implication of parallel evolutionary paths for the two genes (mcr-1 and lptA). Additionally, the functional impairment of the MCR-1 colistin resistance by the removal of the trans-membrane regions demonstrates that membrane anchoring of the soluble catalytic domain (PEA transferase) is essential its function (Table 2). In particular, we also determined the requirement for the five motif-forming residues for MCR-1 function (Table 2 and Fig 4), which might facilitate the binding of this enzyme to its cofactor of zinc ions. The mechanistic insights we obtained definitely extended our knowledge on MCR-1/colistin resistance, and might provide a molecular basis for the development of inhibitors/drugs of small molecule via bypassing the MCR-1-mediated colistin resistance in the post-antibiotics era.

15Generally, the resistance to colistin is correlated with the decrement in affinity of the lipid A group of lipopolysaccharide to the polymyxin antibiotics. Unlike the chromosome-encoded mechanism with the involvement of a two-component systems (pmrAB and phoPQ) and the regulator mgrB, the plasmid-borne MCR-1 represents an newly-emerging machinery for colistin resistance in which the modification of lipid A is catalyzed by the MCR-1 enzyme, giving the reduced affinity to polymyxin (Fig 1). It seems likely that the current situation of MCR-1 colistin resistance worldwide has been over-underestimated since almost 30 countries have been identified to have the mcr-1 gene present in the past several months. Given the fact that i) colistin is the last line of refuge amongst therapeutics against lethal infections by multidrug-resistant Gram-negative pathogens; and ii) the extensive consumption of colistin as a veterinary medicine in the poultry/swine production worldwide functions as a strong selective pressure which then imposes a risky burden on food safety and public health, it is urgently needed to reconsider appropriate use of colistin in veterinary/human medicine and restrict global dissemination of the mcr-1 colistin resistance gene transferred by diversified plasmids. In summary, our findings provide a functional glimpse of plasmid transfer, evolutional origin, and catalysis mechanism for the MCR-1 colistin resistance.

16The clinical E. coli strains from gut microbiota of diarrhea patients (kindly provided by Shenzhen Centre for Diseases Control, China) were grown in the liquid Luria-Bertani (LB) media for the isolation of mcr-1-positive plasmids. The two genetically-modified strains DH5α and BL21 (DE3) were separately applied for gene cloning and protein expression (S1 Table). The colistin-susceptible strain of E. coli MG1655 was used for functional assays for the mcr-1 gene and/or its mutants (S1 Table). The solid LB agar plates supplemented with appropriate antibiotics were applied to either screen possible positive clones for the presence of the mcr-1 gene or determine the minimum inhibitory concentration of colistin by expression of MCR-1.

17The plasmids were isolated routinely from E. coli strains using an alkaline lysis method. Using specific primers (S2 Table), PCR screening was performed for the presence of mcr-1 gene in the colistin-resistant strains. The full coding sequence of mcr-1 was then cloned in pET28(a) via the two cuts (BamHI plus XhoI), giving the recombinant plasmid pET28::mcr-1 (S1 Table). Both the wild type of mcr-1 and its deletion mutant, mcr-1(Δtm) were directly inserted into the two cuts (EcoRI and SalI) of an arabinose-inducible expression vector pBAD24, giving the plasmids of pBAD24::mcr-1 and pBAD24::mcr-1(Δtm), respectively (S1 Table). Similarly, the recombinant plasmid pBAD24::lptA was constructed through cloning of the Neisseria LptA-encoding gene into pBAD24 (S1 Table). Using the pBAD24::mcr-1 plasmid as the template, The experiments of site-directed PCR mutagenesis were conducted as we earlier described. All the acquired plasmids were verified by PCR assays and direct DNA sequencing.

18The minimal inhibitory concentration (MIC) of colistin was determined using liquid broth dilution test as recommended by EUCAST with Cation-adjusted Mueller-Hinton Broth. And survival ability of E. coli expressing different protein was determined as follows: mid-log phase cultures diluted appropriately were spotted on LBA plates supplemented with colistin at varied level (ranging from 0, 0.5, 1.0, 2.0, 4.0, 8.0, 16.0 to 32.0 mg/L), and maintained at 37°C overnight. In the assays for colistin resistance/tolerance, the colistin-susceptible strain MG1655 acted as the negative control, the MG1655 strain carrying the empty vector, pBAD24, referred to blank control, and the MG1655 strain with pBAD24::mcr-1 is the positive control. All the other strains that expressed either Neisseria LptA, the trans-membrane deletion mutant of mcr-1(Δtm), or one of and five mutants of mcr-1 with single point mutation, were tested for their different ability of colistin resistance. 0.2% arabinose was added into the LBA plates to induce the expression of lptA/mcr-1 (and/or its mutants).

19Ultrapure LPS was extracted using the hot phenol method as described by two different groups with minor modifications. Briefly, overnight E. coli cultures (~10 ml) collected by centrifugation were washed with 5 ml of 50% cold acetone before re-suspending in 0.55 ml of water at 70°C. It was then mixed with 0.45ml of phenol (pre-warmed to 70°C) by vigorous vortexing. This mixture was incubated at 70°C for 30 min before spinning at 16,000x g for 15 min to collect the aqueous phase. 1.3 ml of cold absolute ethanol, 6.7 μl of 3 M sodium-acetate and water were added till a final volume of 1.9 ml. This was incubated at -80°C for 15mins to precipitate the crude LPS.

20The resultant crude LPS from the aqueous phase was dialyzed against de-ionized water using the aqueous phase 1000 MWCO dialysis tubing. After dialysis, the samples were freeze-dried, and re-suspended in 55 μl wash solution (20 mM Tris-HCl (pH 8.0), 2 mM MgCl2, DNase I (100 μl of 7 mg/ml) and RNase A (100 μl of 17 mg/ml)). The mixture was maintained for 3 h at 37°C prior to adding 5μl of Proteinase K and further incubating it at 56°C for 1hr. An equal amount of phenol was mixed with the mixture and centrifuged for 30mins at 16,000x g to collect the aqueous phase. 193 μl of 50 mM TRIS, 7 μl of 3M sodium acetate (pH 5.2) and 3 volumes of cold ethanol were added to the aqueous phase and incubated at -80°C for 15mins to precipitate the LPS. The precipitated LPS was collected by centrifugation at 16,000x g for 15 mins and re-suspended in 50 μl water. Finally, Lipid A samples were assayed with electrospray ionisation mass spectrometry (ESI-MS) as Liu et al. reported.

21The mcr-1-positive plasmids that met the requirement of quality control were subjected to library preparation prior to the whole genome sequencing. The next-generation Illumina MiSeq sequencing was conducted as per protocols recommended by the manufacturer, generating a pool of 350-bp paired-end reads. The draft assembly of plasmids was performed with GS De Novo Assembler to give two long contigs. PCR and Sanger sequencing were then performed to verify and correct the contigs. As a result, full genomes of the two plasmids of clinical origins (pE15004 and pE15017) were produced. The plasmid sequences were annotated by RAST, and the genome maps were drawn with the Circos program. Comparative genomics of plasmids were carried out with the tools,Glimmer and BLAST, to probe the potential origin/mechanism for transfer of the mcr-1-carring plasmids.

22To identify sequences homologous to MCR-1, a BLASTp search was performed using the amino acid sequence of MCR-1 and Neisseria gonorrhoeae LptA (formerly named EptA) as a query. In order to avoid hits from very closely related species, Escherichia coli and uncultured environmental samples were excluded from the search and the max target sequences acquired were 500. The top unique protein sequences were selected and submitted to the web-based program, Guidance (http://guidance.tau.ac.il), to evaluate the quality of alignment and to identify potential regions and sequences reducing the quality of alignment. Multiple sequence alignment was performed using MUSCLE with default parameters (http://www.ebi.ac.uk/Tools/msa/muscle/). The alignment was also manually evaluated and adjustments were made as necessary.

23The best amino acid substitution model to be used for reconstructing a tree was identified using the Models function in MEGA version 6. The model with the least score (LG with G+I) was used to reconstruct Maximum Likelihood trees while treating gaps/missing data as partial deletions. Results were validated using 1000 bootstrap replicates.

24Philius Transmembrane Prediction Server (http://www.yeastrc.org/philius/pages/philius/runPhilius.jsp) was applied to probe the topological structure of the MCR-1 protein. The protein sequences of MCR-1 and the related proteins were subjected to the program of Clustal Omega (http://www.ebi.ac.uk/Tools/msa/clustalo/), and the final output of the multiple sequence alignments was given processed by the program ESPript 2.2 http://espript.ibcp.fr/ESPript/cgi-bin/ESPript.cgi). Structural modeling for the PEA-lipid A transferase domain of MCR-1 was processed by Swiss-Model program, using the Neisseria Lipo-oligosaccharide Phosphoethanolamine Transferase A (LptA) as structural template (PDB: 4KAV), the resultant result in ribbon structure was given via PyMol software. Plasmid typing was performed with the help of PlasmidFinder-1.3 server (https://cge.cbs.dtu.dk/services/PlasmidFinder/).

25The genome sequences of the two plasmids (pE15004 & pE15017) were deposited into the GenBank database with the accession no., KX772777 and KX772778, respectively.